使用fastjson读取超巨json文件引起的GC问题
项目中需要将巨量数据生成的json文件解析,并写入数据库,使用了 alibaba 的 fastjson,在实践过程中遇到了 GC 问题,记录如下:
数据大约为70万条,文件大小在3~4G左右,使用 fastjson 官方推荐的 Stream Api 例3 的示例,在读取到30万数据时,内存使用量开始迅速上升,CPU也迅速达到百分之百,在读取到40万数据左右时,出现 GC。
代码如下:
import com.alibaba.fastjson.JSONObject; import com.alibaba.fastjson.JSONReader; import lombok.extern.slf4j.Slf4j; import org.apache.commons.lang3.StringUtils; import org.springframework.beans.factory.annotation.Autowired; import org.springframework.jdbc.core.namedparam.NamedParameterJdbcTemplate; import org.springframework.jdbc.core.namedparam.SqlParameterSourceUtils; import org.springframework.stereotype.Component; import java.io.*; import java.util.*; @Component @Slf4j public class EnterDatabaseUtils { @Autowired private NamedParameterJdbcTemplate namedParameterJdbcTemplate; private final int batchTotal = 50000; public boolean enterData(String databaseName, String tableName, File file, String[] fields) { String fileName = file.getName(); try { JSONReader reader = new JSONReader(new InputStreamReader(new FileInputStream(file.getAbsoluteFile()),"UTF-8")); String insertSql = "INSERT INTO `" + databaseName + "`.`" + tableName + "`" + " (`" + StringUtils.join(fields, "`,`") + "`)" + " VALUES(:" + StringUtils.join(fields, ",:") + ")"; long count = 1; ArrayList<Map<String, Object>> recordList = new ArrayList<>(); reader.startArray(); while (reader.hasNext()) { reader.startObject(); JSONObject = reader.readObject(JSONObject.class); if (count <= batchTotal) { recordList.add(record); count ++; } if (batchTotal + 1 == count) { namedParameterJdbcTemplate.batchUpdate(insertSql, SqlParameterSourceUtils.createBatch(recordList)); count = 1; recordList.clear(); } } if (recordList.size() > 0) { namedParameterJdbcTemplate.batchUpdate(insertSql, SqlParameterSourceUtils.createBatch(recordList)); recordList.clear(); } reader.endArray(); reader.close(); return true; } catch (Exception e) { log.error(databaseName + "." + tableName + ":插入失败"); log.error("", e); return false; } } }
测试代码:
import org.junit.Test; import org.junit.runner.RunWith; import org.springframework.beans.factory.annotation.Autowired; import org.springframework.boot.test.context.SpringBootTest; import org.springframework.test.context.junit4.SpringRunner; import java.io.File; @RunWith(SpringRunner.class) @SpringBootTest public class EnterDatabaseUtilsTest { @Autowired private EnterDatabaseUtils enterDatabaseUtils; @Test public void testEnterDatabase() { File file = new File("/xxx/xxx/xxx.json"); String[] fields = {........}; boolean res = enterDatabaseUtils.enterData("xxxx", "xxxx", file, ); } }
开始的时候,怀疑是 namedParameterJdbcTemplate 引起的内存占用疯涨。但是将所有的数据库相关操作删除,仅保留json读取代码,内存仍然疯涨并导致 GC。
遂怀疑是 fastjson 使用不当,阅读了大量文章之后,终于在 Json少量数据解析 一文中找到了答案:单行直接 readObject 会导致内存不断被消耗!
将代码改为使用 startObject 将每行中的 key、value 单独解析,内存和CPU占用稳定无增长,问题解决。
改进后的代码如下:
import com.alibaba.fastjson.JSONObject; import com.alibaba.fastjson.JSONReader; import lombok.extern.slf4j.Slf4j; import org.springframework.beans.factory.annotation.Autowired; import org.springframework.jdbc.core.namedparam.NamedParameterJdbcTemplate; import org.springframework.jdbc.core.namedparam.SqlParameterSourceUtils; import org.springframework.stereotype.Component; import java.io.*; import java.util.*; @Component @Slf4j public class EnterDatabaseUtils { @Autowired private NamedParameterJdbcTemplate namedParameterJdbcTemplate; private final int batchTotal = 50000; public boolean enterData(String databaseName, String tableName, File file, String[] fields) { String fileName = file.getName(); try { JSONReader reader = new JSONReader(new InputStreamReader(new FileInputStream(file.getAbsoluteFile()),"UTF-8")); String insertSql = "INSERT INTO `" + databaseName + "`.`" + tableName + "`" + " (`" + StringUtils.join(fields, "`,`") + "`)" + " VALUES(:" + StringUtils.join(fields, ",:") + ")"; long count = 1; ArrayList<Map<String, Object>> recordList = new ArrayList<>(); Map<String, Object> record = new HashMap<>(); reader.startArray(); while (reader.hasNext()) { reader.startObject(); while (reader.hasNext()) { record.put(reader.readString(), reader.readObject()); } reader.endObject(); if (count <= batchTotal) { recordList.add(record); count ++; } if (batchTotal + 1 == count) { namedParameterJdbcTemplate.batchUpdate(insertSql, SqlParameterSourceUtils.createBatch(recordList)); count = 1; recordList.clear(); } } if (recordList.size() > 0) { namedParameterJdbcTemplate.batchUpdate(insertSql, SqlParameterSourceUtils.createBatch(recordList)); recordList.clear(); } reader.endArray(); reader.close(); return true; } catch (Exception e) { log.error(databaseName + "." + tableName + ":插入失败"); log.error("", e); return false; } } }
相关文章